Religion and Partisan-Ideological Sorting,

Transcription

1 Religion and Partisan-Ideological Sorting, Nicholas T. Davis March 9, 2017 ABSTRACT This manuscript explores how religious identity, practice, and their joint relationship affect whether individuals connect their ideological to partisan identities a process termed partisan-ideological sorting. Using data from the American National Election Studies (ANES) Time-Series surveys, I find that religiosity constrains the convergence between citizens political identities, with one important caveat: religious identities, and, in particular, evangelical identities, function as the conduit through which religious practice and belief shapes this sorting. Building on these results, I conclude by leveraging the Youth-Parent Socialization (YPS) panel study to estimate the direct impact of religion on sorting over time within a cohort of Americans. Taken together, these findings contribute an alternative social explanation for sorting that complements extant institutional ones. Manuscript word count (inclusive tables, figures, references): 7,303 Key words: identity sorting, ideology, religiosity, evangelicals PhD Candidate, Department of Political Science, Louisiana State University, ndavi21@lsu.edu, nicholastdavis.com 0

2 In a popular government, politics are an important part of religion. No one can possibly be benevolent or religious, to the full extent of his obligations, without concerning himself, to a greater or less extent, with the affairs of human government. - Charles Grandison Finney, I m a Christian, a conservative, and a Republican in that order. ~ Vice President Mike Pence 2 1 Introduction American Christians became increasingly politicized throughout the 1970s and 1980s (Noll, 2002). The idea that religion and politics don t mix was invented by the Devil to keep Christians from running their own country exhorted prominent televangelist Jerry Falwell during a July 4, 1976 sermon, calling on congregants to bend their collective will towards shaping electoral and policy outcomes. And indeed they did. Over the previous four decades, evangelical Christians have played a prominent role in transforming the modern political landscape (see Smidt, Kellstedt, and Guth [2017] for a recent review). In part, these changes were spurred by the comingling of religious and political identification (Steensland and Wright, 2014). Curiously, however, while extant work connects religion to political and ideological affiliation separately (e.g. McDaniel and Ellison, 2008; Patrikios, 2008; Ellis and Stimson, 2012), little research has explored how the nexus of religious beliefs, affiliation, and practice shape the relationship between Americans political identities. Given that this sorting has produced significant levels of affective polarization (Mason, 2015) and behavioral rigidity within the voting 1 Finney, Charles Grandison Lectures on Systematic Theology, Embracing Moral Government, The Atonement, Moral and Physical Depravity, Natural, Moral and Gracious Ability, Repentance, Faith, Justification, Sanctification, & c. Reprint in Lectures on Systematic Theology, vol. 1. By Richard Friedrich. Fairfax: Xulon Press, Republican National Convention Acceptance Speech. Full transcript available at: 1

3 booth (Davis and Mason, 2016), we might ask: what is the relationship between religion and individuals propensity to connect their ideological and partisan identities? This manuscript explores this question using data from the American National Election Studies (ANES) Time-Series surveys covering the period 1984 to I find that religiosity, a reflection of religious preferences that includes cognitive, behavioral, and affective affinities, produces modest, but differentiated effects on identity sorting. Whereas high levels of religiosity translate into higher levels of identity sorting among those with-right leaning political identities, low levels of religiosity which ostensibly convey areligious preferences, if not a form of secularism bear little relationship to sorting among those with left-leaning ones. These results, however, come with an important caveat: the effect of religiosity on sorting flows mainly through religious identity. Although religiosity contributes to significant partisan-ideological sorting among Evangelicals, and to a lesser degree, among mainline Protestants, religiosity has virtually no effect on sorting among Catholic, Jewish, and secular identifiers. Taken together, these findings indicate that the relationship among religious identity, religiosity, and sorting is textured. While both facets of religious expression contribute to greater rates of sorting, the strength of this relationship depends on the underlying nature of an individual s religious identity (or lack thereof), which functions as the conduit through which practice and belief constrain political identities. These findings provide an alternative and compelling social explanation for sorting that complements institutional ones. 2 The politicization of American Christianity While the merging of Christianity and politics appears to be an inescapable feature of the contemporary American political landscape, the strength of this relationship has ebbed and flowed considerably over the course of American history. Jefferson s declaration that the Bill of Rights would separate the church and state like a wall, for example, stands in sharp contrast to de Tocqueville s remarks that for the Americans the ideas of Christianity and liberty are so completely mingled that it is now almost 2

4 impossible to get them to conceive of the one without the other. 3 This disconnect notwithstanding, the ideological, much less partisan, consequences of religion were largely dormant until the mid-1950s, and it was not until the mid-1980s that a politically-conscious Christianity was identifiable, characterized by an extraordinary level of political advocacy to defend traditional positions on abortion, marriage, and religious freedom (Pelz and Smidt, 2015, pg 382). 4 Although the Moral Majority s influence showed some initial signs of faltering in the 1984 election and played almost no role in the 1988 election (Wilcox, 1992), the trend of increasing evangelical support for Republican causes and candidates has held over time. Evangelical support for Republican presidential candidates grew as high as 79 percent in 2012 (Steensland and Goff, 2014), and even Donald Trump, a man not known for public displays of religious piety, gained overwhelming support among Christians during the 2016 election (Pew, 2016). To what source can we attribute this political awakening and conversion? Elitedriven models of group incorporation (e.g. Karol, 2009) and opinion change (e.g. Hillygus and Shields, 2008) suggest that entrepreneurial elites will shift positions to attract or capture new or unaffiliated constituencies. Given that evangelicals were roughly split between the Democratic and Republican Parties in the early 1960s, they comprised something of a sleeping giant of a constituency (Menendez, 1977). According to this narrative, then, we should observe strategic political elites reaching out to conservative evangelicals. 5 In fact, some evidence appears to bear this pattern 3 Alexis de Tocqueville, Democracy in America, ed. by J. P. Mayer. Trans. George Lawrence. 1969, pg Garden City: Anchor Books. 4 Although the First Great Awakening fostered the belief that God favored the revolutionary cause (Marsden, 1990), this produced minimal political mobilization among the religious toward either the parties or the fledgling government. Why? One answer lies in theological content of the period s religious revivals, which emphasized individualist attitudes that challenged such fealty to earthly institutions (Wood, 1993). 5 Consider that Reagan heavily courted evangelicals. Once elected, he was instrumental in changing the official GOP stance on abortion from moderate to pro-life (Karol, 2009) and selected prominent leaders within the Moral Majority s ranks to serve in government positions ranging from the Department of Education to the Surgeon General himself (Cirtchlow, 2007). 3

5 out. For example, Ronald Reagan s spoke to Christian leaders in 1980 affirming the belief that American needed to return to her status as a shining light for the rest of the world I know you can t endorse me but I want you to know that I endorse you and what you are doing! These words portended GOP leaders willingness to accommodate evangelicals and had a transformative effect on religious voters who were previously and weakly affiliated partisans (Williams, 2012; Miller, 2014). Conversely, bottom-up theories of mass realignment (e.g. Carmines and Stimson, 1989; Layman, 2001) convey that group incorporation the particization of a latent constituency occurs when issue evolution creates new dimensions of political conflict that cut across existing political cleavages (e.g. Stoll, 2013). Throughout the 1960s and 1970s, the partisan transformation of evangelicals drew prominently upon culture wars imagery, which pitted the forces of an evangelical worldview against the surging, liberal foe of secular reason (Schaeffer, 1976). As Steensland and Wright (2014, 707) note, framing this tension in these terms allowed evangelicals and their political allies to unite disparate issues such as abortion, gay rights, and women s equality as different manifestations of the same ominous and overarching foe: secular humanism. But there were also prominent economic (Critchlow, 2007) and racial (Noll, 2008) concerns that motivated this transformation. Not only did the fundamentalistmodernist split intersect the class-based cleavages borne from the New Deal era (McTague and Layman, 2009), but the passage of civil rights legislation freed Southern, religiously conservative whites from the Democratic Party. 6 In tandem with 6 The role of race in this conversion should not be understated. In fact, the theological divide that characterized the Northern schism between fundamentalists and modernist evangelicals did not as readily apply to the South, where such divisions essentially fractured along racial lines (see Wadsworth, 2014 for a more detailed treatment of this thesis). Noll (2008, ) considers how the mobilization of Southern whites was partially attributable to these changes: [O]nce legally enforced racism was gone, the great impediment that had restricted the influence of southern religion to only the south was also gone. Stripped of racist overtones, southern evangelical religion the preaching, the piety, the sensibilities, and above all the music became much easier to export throughout the country. 4

6 enterprising, evangelical organizations (Williams, 2012), the grafting of religious conservatives to the Republican Party drew from grassroots origins. It s possible, however, that these two explanations are not mutually-exclusive and, instead, are symbiotic. Indeed, if we allow both party elites and evangelicals some degree of strategic agency, then the relationship might be reciprocal (McTague and Layman, 2009; Williams, 2012). Oldfield (1996) notes, for example, that activist groups make strategic calculations regarding what course of action will satisfy their political interests. The probability that a group will become involved with a particular party, then, depends on both party elites relationship to the issues that the activist group holds dear and the course of action that group leaders pursue to produce their optimum policy preferences. Republican Party elites, for example, incorporated the language of civil religion into their platform in the early 1980s, which affirmed religious conservatives belief in the religious nature of the founding. In turn, evangelicals pushed the GOP further rightward to accommodate their bundle of social policy preferences (Noll, 2002). Thus, the fact that evangelical leaders, activists, and voters became fiercely loyal to the GOP and highly influential within the party may be due just as much to the strategic calculations of Christian Right leaders and activists as it is to those of Republican politicians (McTague and Layman, 2009, 344). Put another way, If evangelical Christians had become Republicans, the Republican Party had also become Christianized (Williams, 2012, pg. 231). 3 Religion and the sorting of political identities This behavioral sorting implies the existence of a broader ideological and partisan realignment among Christians. There are at least two mechanisms by which religion might affect identity sorting, which constitutes the convergence between partisanship and liberal-conservative identification. First, prior research reveals a strong relationship between elite cues and sorting (Levendusky, 2009; Davis and Dunaway, 2016). Although this work focuses primarily upon how elite polarization helps individuals connect their ideological to partisan preferences, a similar cue-taking mechanism is embodied in clergy and pastors exhortations connecting religion and 5

7 politics (Williams, 2012). Yet, while it is likely that religious elites contribute to sorting by explicitly connecting ideological to partisan preferences from the pulpit, even the mere act of worship or practice within a religious tradition might contribute to sorting inasmuch that church attendance or scriptural beliefs, for example, provide information (heuristics) that may constrain political preferences. In this way, various components of religious experience and practice cue individuals to the correct political preferences. 7 Second, consider the motivational qualities that underscore sorting. Individuals tend to avoid cognitive dissonance among their preferences (Festinger, 1957). Past research, for example, shows that citizens acquire positive candidate impressions that correspond to their electoral choices (Caplan, 2001) and harmonize their political attitudes accordingly (Mullainathan and Washington, 2009; McGregor, 2013). Simply put, people tend to deal with attitudinal and behavioral discrepancies because these incongruences arouse psychological discomfort (see Crano and Prislin, 2006, for a review). If political ideology originates from a range of dispositional motives (e.g. Jost et al., 2003) motives that also underscore religious inclinations (e.g. Jost et al., 2008; Firesen and Ksiazkiewicz, 2015) then the relationship between political and religious affinities may be sufficiently strong to compel individuals to align their associated political identities to reduce the dissonance among them. Thus, a lack of discordance among religious and political preferences would reduce the type of cross-pressures that lead to conflicted identities (Campbell and Miller, 1957; Roccas and Brewer, 2002). Taken together, these mechanisms suggest a strong link between religion and sorting. Yet, while extant work connects the politicization of Christianity to both ideological and partisan identification (e.g. Layman and Hussey, 2007; Ellis and Stimson, 2012; Schwadel, 2017), little empirical research has examined how religion 7 There is some question of temporality here. Patrikios (2008) and, more recently, Goren and Chapp (2016) take the position that partisanship and public opinion may shape religious preferences and orientations. While the panel data needed to adjudicate temporality is effectively nonexistent, I address this concern both theoretically and empirically later in the manuscript. 6

8 shapes the convergence between political identities. Indeed, while Levendusky (2009) offers qualified evidence that evangelical identification increases individuals propensity to correctly match their ideological preferences to their partisanship, there are two problematic aspects with this analysis. First, he does not discriminate between policy and identity sorting, which are distinct dimensions of political sorting (blinded). Second, although Levendusky finds that evangelicals are more likely to be sorted, the role of religiosity perhaps the most important indicator of religious belief and practice (Hackett and Lindsay, 2008) is left completely unexplored. What is missing from this extant research, then, is a detailed, empirical accounting of how religious belief, practice, and affiliation affects citizens propensity to match their partisan to ideological identities. For our purposes, two concepts tap into different facets of religion that might shape partisan-ideological sorting: 1) religiosity, which is generally understood as a matrix of attitudes, behavior, and values (Glock, 1965; Lenski, 1963), and 2) religious or denominational affiliation, which is more reflective of a social identity (Gallagher, 2004; Huddy, 2013; Smidt, Kellstedt, and Guth, 2017). Beginning, first, with religiosity, the sociological expression of this concept often combines participation, saliency, and belief acceptance into a latent measure of orthodoxy (Cardwell, 1980; Holdcroft, 2006). In effect, this concept yields an index that juxtaposes areligious (and, perhaps secular ) preferences with conservative or fundamentalist religious orthodoxy. Thus, religiosity captures the type of ideational content that might constrain political identities, with one caveat: because sorting is a measure of the integration between political identities i.e. it has no inherent left-right valence we ought to expect that the effect of religiosity on sorting should be positive for those persons with right-leaning identities and negative for those with left-leaning ones. If religiosity reflects a vague left-right dimension of secular-fundamentalist religious preferences, then it is likely that it should exert different effects for persons with left- and right-leaning political identities. 7

9 H1: Religiosity should increase sorting among individuals with right- but not left-leaning identities. Second, there is an important distinction to be made between the measure of religiosity and religious identity. Religious identity resembles other social identities in that they generate political cohesion through a common or shared group perspective and conformity to associated group norms (Huddy, 2013; Miller et al., 1981; Simon and Klandermans, 2001). Because social identities can bind individuals to goal-directed behaviors (thereby removing troubling cross-pressures), it is likely that religious identity should constrain concomitant political identities. Given prior work which shows a connection between evangelicalism and republican identification (e.g. McDaniel and Ellison, 2008) and ideological preferences separately (e.g. Ellis and Simson, 2012; Farizo et al., 2016), I expect that evangelical identification should produce greater partisan-ideological sorting relative identification with other (or no) faith traditions. In fact, it may be the case that the effect of religiosity only funnels through particular religious identification as an evangelical. Given the close correspondence between traditional evangelicalism and the type of fundamentalist orthodoxy that religiosity captures, self-identified evangelicals who espouse high levels of religiosity should be particularly likely to possess sorted political identities. H2: The magnitude of the effect of religiosity on sorting should be most pronounced among Evangelicals. 4 Data and Measurement The data utilized in the forthcoming analyses are drawn from two sources: The American National Election Studies (ANES) Time-Series and the third (1982) and fourth waves (1997) of the Youth-Parent Socialization surveys. 8 Across both 8 Unfortunately, while the 2016 ANES Time-Series was released in late March, the coding for a number of key items (e.g. partisanship, church attendance, etc.) contains either missing data or raw responses that have yet to be coded. 8

10 datasets, relevant covariates are coded consistently and every effort has been made to ensure model specifications are comparable. 4.1 Operationalizing identity sorting A growing body of research conveys that partisan (Greene, 2002) and liberalconservative identities (Malka and Llekes, 2010) can be conceptualized as forms of social identities. Thus, partisan-ideological sorting is a concept that conveys the extent to which an individual s separate political identities cohere. Following past research (e.g. Mason, 2015; Davis and Mason, 2016), this identity sorting is operationalized by creating an identity alignment score between the traditional seven-point partisan and ideology scales, where low values convey Democrat / liberal identification and high values Republican / conservative identification. Taking the absolute difference between these two items produces a variable where low values convey perfect overlap between identities and high values weak and cross-pressured ones. This variable is reverse-coded so that the maximum value represents a perfect ideology-partisan identity match and then multiplied by folded measures of partisan and ideological strength. I then rescale the resulting item so that sorting values range along a 0 to 1 continuum, where the least-aligned, weakest identities are coded 0 and the most aligned, strongest identities are coded Religiosity and Denominational Affiliation A substantial literature is devoted to the study of how religion is both theoretically and empirically operationalized (Smidt, Kellstadt, and Guth [2017] for a recent review). Religiosity, in particular, is a thorny concept in that it embodies a number of broad constructs. As Holdcroft (2006, 89) notes, religiosity is colloquially synonymous with such terms as religiousness, orthodoxy, faith, belief, piousness, devotion, and holiness. Further, because the concept transcends a number of academic disciplines, which seem to approach the study of religion from qualitatively and stylistically different vantage 9

11 points (Cardwell, 1980), it can be difficult to find concerted theoretical agreement regarding the empirical expression of this concept. 9 For the purposes of this analysis, religiosity is defined using a factor model of three items that share lengthy iterative histories across the ANES Time-Series surveys: biblical literalism, church attendance, and self-professed importance of religion. Biblical literalism asks respondents whether they believe that the Bible is a mere book written by men (coded 0), inspired by God but not to be taken literally (coded 1), or the literal Word of God (coded 2). Self-reported church attendance ranges from 0 to 4, where a score of 0 corresponds to never attending church, while the maximum value conveys that an individual attends church multiple times a week. 10 Importance of religion is coded 1 for yes, religion is important and 0 for no, religion is not important. These items are subjected to an iterated principle factor analysis. The factor loadings range from 0.75 for bible, 0.78 for church attendance, and 0.80 for religion is important, and the variables cleanly load onto a single common factor (first extracted eigenvalue = 1.81, second = 0.63). To interpret the conceptual nature of this item, it helps to consider how these factor scores relate to different combinations of theses constituent variables, which are coded in such a way that low values roughly approximate liberal responses perspectives and high values conservative ones. In this case, a maximally-positive religiosity score would represent a combinatory matrix of items that convey an individual attends church regularly, believes the Bible is the 9 Lenski (1963), for example, portrays religiosity as a combination of associational, communal, doctrinal, and devotional facets. Although Glock and Stark (1965) pursue a somewhat similar operationalization strategy, they use different terminology and expand this definition to include experiential, ritualistic, ideological, intellectual, and consequential dimensions. Meanwhile, Fukuyama s (1960) interpretation of religiosity diverges prominently from these approaches and instead comprises a combination of cognitive, cultic, creedal, and devotional qualities. Still other research argues that these multidimensional approaches problematically aggregate orthodoxy, or religious orientation, and practice, or religious commitment, together (Davidson and Knudsen, 1977). 10 These scores are probably inflated as a function of social desirability effects, but the problem, unfortunately, cannot be readily corrected across either survey (Hadaway et al., 1993; Patrikios, 2008). 10

12 inerrant Word of God, and is a conservative fundamentalist. Low scores, conversely, would represent church non-attenders, those who believe the bible is a mere book of stories, those who are religious liberals, and don t believe in God. Substantively, then, this measure of religiosity effectively ranges from a secular to more fundamentalist approach to religion. Separate from the concept of religiosity is the measurement of religious identity. This concept comprises an individual s self-identified denominational affiliation (e.g. Glock and Stark, 1965), and although my primary interest is concerned with the measurement of evangelical identity, I also utilize identification with many of the major faith traditions below. The operationalization of evangelicalism presents its own peculiar difficulties (see Hackett and Lindsay [2008] for an excellent review). For instance, evangelicals may espouse particularistic beliefs about salvation, sanctification, and the devil (Barna, 1994), yet they also may literally self-identify as evangelicals when given the option. Complicating this measurement scheme is a two-pronged measurement strategy that treats the status of being born-again, a question first asked by Gallup in the mid-1970s (Hackett and Lindsay, 2008) an important spiritual marker that communicates that an individual has expressed belief in the redemptive power of Jesus Christ and now possesses a personal relationship with him (Smith, 2000) as functionally equivalent to explicit identification as an evangelical. This latter indicator is perhaps the simplest general criteria that demarcates evangelicals from other Protestants (e.g. Steensland et al., 2000), and it avoids entangling group affiliation with other belief-aspects that may be wrapped up in a measure of religiosity (Hackett and Lindsay, 2008). Moreover, it is the longest running indicator of evangelical identity within the ANES survey; questions that explicitly asked whether respondents identify as an evangelical were only surveyed for a brief period of time through the later 1980s. To operationalize evangelical identification, I assign the value 1 to respondents who identify as both Protestant (the parent religious 11

13 category) and born-again, and otherwise Denominational affiliation is further broken down into the following mutually-exclusive categories that roughly follow Steensland and colleagues (2000) recommended taxonomy: Mainline Protestants (all Christians who do not convey that they are born-again ), Catholics, those persons who identify as Jewish, and the excluded category Secular / other faith identifiers. 12 Individuals are coded 1 for identifying with a particular group and otherwise Additional covariates A number of controls are also employed in the forthcoming analyses. Political interest ranges from 0, low interest to 2 high interest. Knowledge of House majority is used as a proxy for political knowledge, where correctly identifying which party controls the House of Representatives during the previous election cycle is coded 1 and incorrect answers are coded Education is measured as an ordinal variable that ranges from 0, primary, to 6, post-graduate. Age takes the form of a continuous variable that ranges from 17 to 99. Male is coded 1 for men and 0 for women. White and Black racial identification is coded 1 for individuals who select those categories and otherwise 0. Old South is a variable utilized in studies that examine sorting over time to control for the effects of southern realignment. Individuals residing in one of the states that comprised the original Confederacy are coded 1 and otherwise 0. Finally, a number of alternative variables reflecting potential alternative explanations for sorting are included in the forthcoming analyses. Differentiating between economic and moral policy preferences is a strategy that accounts for recent research that suggests that policy preferences are minimally two-dimensional. Abortion 11 Given that whites overwhelmingly identify as evangelicals across racial groups (see Steensland et al., 2000), this variable is restricted only to white identifiers. 12 Unfortunately the ANES does not discriminate secular or religious nones from others in the cumulative or pooled ANES Time-Series data. 13 Unfortunately, the broader office recognition items that are popularly utilized in later surveys are unavailable for a significant span of the early data (1980s). It is important not to lose this period of data, so this item provides the best conceptual analogue. 12

14 preference captures an individual s self-placement on the four-category abortion selfplacement scale, which ranges from 0, always permissible to 1, never permissible. Economic policy preferences comprise a factor index of four self-placements on government spending, provision of government health insurance, aid to minorities, and the provision of jobs (Eigenvalue = 2.24), where higher values convey a preference for economic conservatism. In addition to policy preferences, perceived polarization is also included in order to test the relative effect of religion on sorting against this common institutional explanation (e.g. Levendusky, 2009; Davis and Dunaway, 2016). 5 Results Table 1 depicts sorting as a function of religiosity, denominational affiliation, and a series of covariates that are shown elsewhere to be related to identity sorting. The results portrayed in the first column indicate that religiosity has a statistically significant, though modest effect on sorting. Compared to average levels of sorting (y = 0.25), partisan-ideological sorting increases by about 20 percent as individuals transition from minimum to maximum values of religiosity. However, there is good reason to suspect that this effect is not uniform for all respondents. In particular, given the close correspondence between religious groups (e.g. the religious Right ) and right-leaning identities (Republicanism, conservatism), it may be the case that the effects of religiosity on sorting vary among how citizens connect partisanship and ideology. Indeed, there is some evidence that religiosity exerts differentiated effects on sorting that are masked when analyzing the full sample. As Model 2 shows, religiosity has a significant, though small in magnitude negative effect on sorting for those with left-leaning identities. However, for individuals with rightleaning identities the magnitude of the effect of religiosity on sorting almost doubles (Model 3). 13

15 Table 1. Effect of religiosity on Sorting Full sample Left-leaning identities Right-leaning identities Religiosity 0.05** -0.02* 0.09** (0.00) (0.00) (0.00) White Evangelical 0.03* (0.01) (0.01) (0.01) Mainline Protestant (0.01) (0.01) (0.01) Jewish * (0.01) (0.01) (0.05) Catholic -0.04** -0.02* -0.04** (0.01) (0.01) (0.01) Political interest (0.03) (0.02) (0.02) Knows House majority 0.03** 0.01* 0.02 (0.01) (0.01) (0.01) Abortion preference 0.06** -0.06** 0.13** (0.01) (0.01) (0.01) Economic policy preferences 0.01** -0.08** 0.09** (0.00) (0.00) (0.01) Perceived polarization 0.35** 0.23** 0.31** (0.02) (0.02) (0.02) Male (0.01) (0.01) (0.01) White 0.03** 0.02** 0.02 (0.01) (0.01) (0.01) Black (0.01) (0.01) (0.05) Age (0.00) (0.00) (0.00) Education 0.01** 0.01* 0.00 (0.00) (0.00) (0.00) Old South -0.02** (0.01) (0.01) (0.01) Constant -0.07* 0.06* -0.06* (0.03) (0.03) (0.03) R N 9,107 4,296 4,036 Source: CANES Time-Series Notes: Models for left- / right-leaning identities are split-samples by PID. Full model includes year fixed effects; see Table A2 in Appendix. Standard errors in parentheses; *p<0.05, **p<

16 The shape of the relationship between sorting and religiosity, however, requires also addressing the issue of time specifically, the dynamism inherent in the merging of religious conservatism and Republicanism (Williams, 2012). Structural cleavages, like religion (but also race, ethnicity, language, etc.) are often exacerbated when a cleavage is explicitly connected to political issues on which party elites stake out distinct positions (McTague and Layman, 2009; Sundquist, 1983; Layman, 2001; Leege et al., 2002) a process that naturally takes time to unfold. If issue orientations precede group images (McTague and Layman, 2009, 353), then the degree to which religion is related to sorting should be sensitive to the historical unfolding of the relationship between Republican Party elites and religiously-conservative constituencies. Empirically, time ought to moderate the effect of religiosity on sorting such that the magnitude of the effect of religiosity increases over the period 1984 to 2012 Rather than constraining the effects of time to a linear interaction model, however, I present in Figure 1 plots of the coefficients for religiosity at individual time points within the ANES Time-Series surveys. Panels A and B reveal two important insights. First, the pooled models of sorting in Table present the average effect of religiosity on sorting at maximum power. Although the coefficient for religiosity pointed to a negative, statistically-significant relationship between religion and sorting, Panel A reveals that the confidence intervals for the effect of religiosity on partisan-ideological sorting at each survey-year overlap with 0.00 (solid dark line). In other words, while we might expect religiosity to undercut sorting for individuals with left-leaning identities, this illustration shows that the effect is more or less non-existent. As we transition to Panel B, it is clear that there are large swings in the association between religiosity and sorting among Republicans. From 1984 to 1992, the marginal effect of religiosity on sorting overlaps with In 1992 and 1996 the coefficient for religiosity is modest, though statistically differentiable from 0.00, although the effect washes out during the 2000 election. However, from 2004 to 2012, the magnitude of the effect of religiosity on sorting is large in 2008, persons who scored at the upper threshold of religiosity where roughly doubled the average sorting score. 15

17 Figure 1. Marginal effect of religiosity on sorting, by left- and right-leaning identities over time Source: CANES Time-Series Notes: Point estimates convey the marginal effect of transitioning from minimum to maximum values of religiosity for respondents with left- (Panel A) and right-leaning identities (Panel B) in given year. Individual yearly models from which estimates are drawn are available in Appendix A. 16

18 Finally, among the various covariates for which I control, the strength of the relationship between sorting and abortion preferences relative sorting and economic preferences is noteworthy. Specifically, the magnitude of the coefficient for abortion attitudes is larger than economic ones, which sheds some additional light on the nature of the relationship between religion and sorting. Individuals with right-leaning identities who take pro-life stances are considerably more sorted than those who hold similar abortion preferences yet identify with the left. Further, this size of this coefficient is slightly larger than even the one for religiosity, which conveys that orthodoxy is not a strict replacement for certain policy orientations with deeply religious connotations. In this way, ideological sorting among Republicans is as much a function of route religiosity as it is a specific derivative of how these individuals think about the abortion. Having shown that the effect of religiosity on sorting varies both among Democrats and Republicans and over time, I turn next to exploring the effect of religiosity on sorting when filtered through religious identities. Table 2 again depicts the relationship among sorting, religiosity, and a series of controls for individuals who identify with varying religious denominations. Here, the effect of religiosity is largely isolated among Evangelicals and Mainline Protestants (secular identifiers are the excluded category for these analyses). The results of these analyses indicate that maximum levels of religiosity convert to almost a 70 percent increase in sorting among Evangelicals and about a 20 percent increase for Mainline Protestants. 17

19 Table 2. The effect of religiosity on sorting by denominational affiliation Evangelical Catholic Mainline protestant Jewish Religiosity 0.17** ** 0.03 (0.01) (0.00) (0.00) (0.01) Political interest (0.02) (0.03) (0.02) (0.05) Knows house majority * 0.12** (0.02) (0.02) (0.01) (0.04) Abortion preference 0.13** * 0.06 (0.02) (0.01) (0.02) (0.09) Economic policy preferences 0.07** ** -0.09** (0.01) (0.01) (0.01) (0.02) Perceived polarization 0.36** 0.30** 0.34** 0.50* (0.03) (0.03) (0.03) (0.18) Male (0.01) (0.01) (0.01) (0.05) White * 0.02* 0.11 (0.01) (0.01) (0.05) Black (0.03) (0.01) (0.18) Age (0.00) (0.00) (0.00) (0.00) Education ** (0.01) (0.00) (0.00) (0.02) Old South -0.03* ** 0.00 (0.01) (0.02) (0.00) (0.05) Constant * (0.03) (0.05) (0.04) (0.07) R N 1,427 2,148 3, Source: CANES Time-Series Notes: Models include yearly fixed effects; see Table A3 in Appendix. Standard errors in parentheses; *p<0.05, **p<

20 Figure 2. The effect of religiosity on sorting, conditional on religious identity Notes: Estimates derived from models presented in Table 2. Y-axis conveys predicted sorting at varying values of religiosity (x-axis). Shaded bands convey 95% confidence intervals. Figure 2 illustrates the shape of the relationship between religious identity and sorting for Mainline Protestants, Catholics, and White Evangelicals. While Catholics receive virtually no change in sorting as religiosity increases, Mainline Protestants are modestly more sorted when religiosity increases from minimum to maximum levels. However, the effect of religiosity on sorting among Evangelicals is even more striking. Evangelical identity exerts little constraint upon partisan-ideological sorting without concomitant, high levels of religiosity. In fact, even at middle values of religiosity, it is not clear that the separate religious identities exert differentiated effects on sorting. At maximum levels of religiosity, though, it is clear that white Evangelicals are more wellsorted than those individuals belonging to other groups, which conveys that the power 19

21 of identities are at least partially contingent upon the substantive (ideational) foundation upon which those identities derive their meaning. Again, given that these effects are drawn from analyses that pool all Time- Series data, it can be difficult to assess how the relationship between religiosity and sorting varies over time, particularly among Evangelicals. To this end, Figure 3 portrays the marginal effect of transitioning from minimum to maximum values on religiosity on sorting for each year of survey data. Note that the magnitude of this effect is quite large initially, decreases to zero during 1988 and 1992 (matching the results of the earlier analysis of the shape of the effect of religiosity on sorting among those with left- and right-leaning identities), and steadily grows from 1996 to 2012, albeit not in a linear fashion. On balance, white Evangelicals who are regular church attenders, believe the bible is the inerrant Word of God, and communicate that their religion is important to them are roughly two times more sorted than the average citizen. The evidence presented thus far reveals that the effects of religiosity on sorting are magnified among those with right-leaning partisan identities and who identify as white Evangelicals. However, as the careful reader may note, perhaps it is the combinatory alignment of these three features a phenomenon sometimes referred to as social sorting (Mason and Davis, 2016) that is responsible for partisan-ideological sorting. In other words, perhaps the effects of religiosity on sorting flow through religious identity. 20

22 Figure 3. The effect of religiosity on sorting among Evangelicals, Source: CANES Time-Series, Notes: Point estimates convey the marginal effect of transitioning from minimum to maximum values of religiosity for Evangelical identifiers in given year. Individual yearly models from which estimates are drawn are available in the Appendix, Table A6. Estimates bracketed by 95% confidence intervals. Figure 4 illustrates a series of illustrations that graphically portray the results of three-way interactions among a three-category partisanship item, religiosity, and religious identity. 14 For Secular, Catholic, and Mainline Protestants, the effect of 14 To focus on the visual impact of these effects, the output for this illustration is relegated to Appendix A, Table A8. By including the three-category partisanship item among right-hand-side variables, we are merely isolating how ideology and partisanship cohere among specific groups. 21

23 religiosity on sorting does not vary by religious identity. Indeed, the x-shaped pattern in each three panels suggests that religiosity exerts a more or less uniform effect among varying denominational affiliations; instead, the effect of religiosity seems to flow primarily through partisan identities insofar as religiosity contributes to greater observed sorting among those with right-leaning identities and less sorting for those with left-leaning ones. In other words, while a lack of religiosity contributes to high levels of sorting among those with left-leaning identities, persons with right-leaning identities who are also highly religious are more likely to be well-sorted. In fact, the only case in which social sorting translates to greater correspondence between political and ideological identities appears in the fourth panel. White Evangelicals, who are Republicans and who possess heightened levels of religiosity are more likely to possess well-sorted partisan and ideological identities. Having demonstrated that religiosity and sorting are closely-related, albeit in a textured manner, one final question with which we might be concerned, however, is whether sorting exerts an effect on religiosity or religiosity on sorting. 15 In other words, what is the ordering of the relationship between these two phenomena? In the absence of panel surveys that routinely poll the same group of individuals, it is difficult to establish causality in the relationship between religion and sorting (e.g. Patrikios, 2008). However, using the Youth-Parent Socialization survey, we can begin to inquire into the nature of this relationship Recent work, for example, notes that culture war attitudes may be responsible for constraining political and religious predispositions (Goren and Chapp, forthcoming), while other research shows partisanship can influence church attendance (Patrikios, 2008) which implies that political considerations affect religious ones, rather than the inverse (so argued here). 16 The variables included in the analyses found in Table 3 are coded as consistently as possible in relation to the ANES Time-Series variables. For a full discussion of these items, Appendix B lists the exact coding and summary statistics of this data. 22

24 Figure 4. The effect of religiosity on sorting, conditional on political and religious identity Source: CANES Time-Series Notes: Estimates depict three-way interaction among partisanship, religiosity, and denominational affiliation. For full model from which estimates are derived, see Table A4 in Appendix A. Estimates bracketed by 95% confidence intervals. 23

25 To explore the interdependence of these two constructs, I estimate panel models that compare the extent to which a variable at T1 predicts the T2 value of a second variable (an approach used previously by Jennings and Stoker [2004]). Specifically, I model religiosity at the fourth wave of the YPS survey (1997) as a function of sorting at the third wave (1982), and sorting at the fourth wave of the YPS survey (1997) as a function of religiosity at the third wave (1982). By controlling for a series of covariates measured at the same wave as the dependent variable (i.e. T2), this set of analyses offers insight into the extent that someone who is, for example, highly religious at the first time point ends up more sorted a the second than would otherwise be expected given their demographic profile, for which I control. 17 Beginning with the first two models presented in Table 3, respondents religiosity scores in 1997 are modeled as a function of Soring 1982 and a series of controls. Drawing from the presentational style in the previous section, which revealed that there were prominent differences in the way that religiosity affects sorting for those with right- and left-leaning identities, there is little evidence that the extent to which a person was sorted in 1982 predicts their religiosity in 1997 as evidence by the large standard errors associated with the coefficients for sorting. However, turning to the second set of models, there does appear to be a modest, but positive relationship between religiosity at 1982 (T1) and sorting in 1997 (T2). Specifically, higher levels of religiosity are related to an increase sorting. While this does not strictly reveal that religiosity causes sorting, it does help alleviate some of the concern about the 17 An alternative approach would be to utilize a structural equation model. SEM, however, requires strong assumptions to identify a model and is sensitive to exclusion restrictions. Although I cannot provide a dynamic model of the causal relationship between religiosity and sorting, this modeling approach leverages panel data collected among individuals who, at this point in their lives, would be undergoing the type of individual-level experience whereby they continue down the path of adolescent religious indoctrination or release themselves from religious practice (for a different application of this same modelling approach, see Jennings and Stoker [2004]). 24

26 directionality of the relationship between religion and sorting in that the effects are not reciprocal. Table 3. Interdependence of religiosity and sorting, Religiosity 1997 Sorting 1997 Left-leaning Right-leaning Left-leaning Right-leaning Sorting (0.07) (0.07) Religiosity ** (0.04) (0.05) Catholic -0.03** -0.02* 0.03** 0.01 (0.01) (0.01) (0.01) (0.01) Jewish (0.03) (0.04) (0.02) (0.04) Southern Baptist -0.13** -0.32** (0.04) (0.10) (0.04) (0.09) Secular (0.04) (0.05) (0.03) (0.05) College degree -0.28** -0.37** 0.09* 0.05 (0.04) (0.06) (0.04) (0.06) White * 0.06** (0.02) (0.03) (0.02) (0.03) Black (0.02) (0.03) (0.02) (0.02) Political knowledge (0.05) (0.22) (0.04) (0.21) Constant 0.83** 0.82** * (0.04) (0.04) (0.05) (0.06) R N Source: Waves 3 (1982) and 4 (1997), Youth-Parent Socialization Survey Notes: All covariates save religiosity and sorting are sampled from 4 th wave (1997). Standard errors in parentheses. *p<0.05, **p<

27 6 Conclusions To what do Americans owe the significant identity sorting that has accrued over time? Conventional explanations of this phenomenon have primarily focused on the effect of elite cues (Levendusky, 2009) and structural changes in the media environment (Davis and Dunaway, 2016). This manuscript, however, shows how religious considerations are also associated with such sorting: religiosity, and, by extension, Evangelical identity, exert a strong positive effect on partisan-ideological sorting. Put another way, the funneling of religiosity through evangelical identity is essentially a form of social sorting that leads to political sorting. Scholars have tracked the reliable relationship between the mass public s religious and political orientations for almost thirty years. Over this duration of time, original bloc of Christian or Religious Right has fractured, undergoing notable changes. Not only have the formal institutional structures that bound religious individuals disappeared (e.g. Cooperman and Edsall, 2006), but a new crop of less conservative religious leaders and pastors have emerged as vocal critics of those fundamentalist groups (e.g. Altman and Scherer, 2012). Combined with the various new items that have moved onto the issue agenda, ranging from environmentalism (Harden, 2005) to racial relations (Wadsworth, 2001), it is conceivable that divisions that originally bound religious conservatives to the Republican Party had fractured enough to lose their purchase. Instead, the results of these analyses show that the same religious orthodoxy or fundamentalism that sorted individuals in the mid-1980s has largely persisted to the present. To be sure, the effects of religiosity on sorting are not ironclad. In fact, during the late 1980s and early 1990s, the effects of religiosity are highly variable, perhaps due in part to the fracturing of the aforementioned institutional structure of the Christian Right. But be that as it may, the evidence presented here shows that religious-social sorting powerfully constrains partisan-ideological sorting, even occasionally approximating the effects of elite cues. These effects, however, are largely 26